DeepSeek: Everything you need to know about the AI chatbot app (techcrunch.com)

DeepSeek, a Chinese AI startup spun out of quant hedge fund High-Flyer, surged to mainstream attention after its chatbot app topped app-store charts worldwide. Founded as an offshoot of High-Flyer’s AI work, DeepSeek built in-house data centers and — despite U.S. export limits that forced it to use Nvidia H800 chips rather than H100s — released a rapid succession of models (DeepSeek V2, V3, V3.2-exp and the R1 reasoning model) that the company says match or beat competitors on standard benchmarks while running much cheaper. Its models are not “open source” in the traditional sense but are available under permissive licenses on Hugging Face, where R1 derivatives have accrued millions of downloads; the chatbot reached tens of millions of visits even as usage dipped from its peak. The story matters because DeepSeek challenges assumptions about AI leadership, compute demand, and business models: efficiency claims have pressured domestic rivals to cut prices, prompted market reactions (including an early hit to Nvidia’s stock), and triggered regulatory and security scrutiny. Technically notable is R1’s reasoning architecture — slower by seconds to minutes but more reliable for science and math — and V3.2-exp’s focus on dramatically lower inference costs for long-context tasks. That combination of claimed efficiency, permissive licensing, and rapid adoption has spurred partnerships (Microsoft Azure listing) and bans (U.S. government devices, some states and countries), raising fresh debates over supply-chain limits, model provenance, and geopolitical risk in AI.